"... which they are willing to share in exchange for a fifty-fifty split of the proceeds. Such a person inadvertently reveals that they think formulating sentences is a nuisance rather than a fundamental part of storytelling in prose. Generative A.I. appeals to people who think they can express themselves in a medium without actually working in that medium. But the creators of traditional novels, paintings, and films are drawn to those art forms because they see the unique expressive potential that each medium affords. It is their eagerness to take full advantage of those potentialities that makes their work satisfying, whether as entertainment or as art...."
Writes Ted Chiang in "Why A.I. Isn’t Going to Make Art/To create a novel or a painting, an artist makes choices that are fundamentally alien to artificial intelligence" (The New Yorker).
Subscribe to:
Post Comments (Atom)
59 comments:
AI won't make great art as perceived by humans. But who says humans are the arbiters of what makes art great? When AI (I prefer ML to AI, for what it's worth) is discussed, I think the qualifier *yet* should always be inserted.
"an artist makes choices that are fundamentally alien to artificial intelligence" That's why we call it artificial. Yes, it doesn't "think," and it doesn't "make choices." But it can mimic thinking and choosing quite well, and more quickly. Leaving aside that many artists make bad choices in the first place.
”Many novelists have had the experience of being approached by someone convinced that they have a great idea for a novel..."
I did not know that. Ted Chiang thinks AI won’t create art? He’s missing the word “yet.”
Successful art is what people will pay for.
Sometimes that's not what the academics predict.
Sometimes it's even what the experts abhor.
Will AI replace the "big" art? No. But it will take the jobs artists used to do while struggling before they made the "big" thing. Few make a living from "big" art, most did commercial and media art to pay the bills. Now they'll need to do plumbing by day, "big" art by night.
Maybe they can manage anthropogenic mimicry, but there is no evidence that machines can innovate.
Even as far back as the 1940s, George Orwell imagined books being written by machine.
AI is a tool. Humans have shown many creative ways to use tools. Why would this be any different?
So this guy ships out on a whaler, with a captain who's obsessed about a white whale. It almost writes itself.
We are going to see a lot of articles like this, because it is a fundamental truth that AI is derivative of existing art and cannot create a new piece of art. It can only alter existing art to generate facsimiles of a style. Creating a new style or novel concept is something only people can do. I commend Althouse for wanting to remain able to distinguish AI content from "real" content. After all, the name gives it away. It's not intelligent, it's artificial intelligence.
Which is why it can be used to replace braindead NPCs in the mainstream media who simply act as stenographers to the DNC. Or it could listen to an NFL/NBA/MLB broadcast and write a summary in the style of wire copy. Not like Bill Plaschke or Dan Shaughnessy, but easily to the UPI wire service stylebook.
Won't create great art, enduring art. Art with heart.
It'll certainly disrupt the James Patterson/Thomas Kincaid business model.
I doubt it will be that long before AI will be capable of turning out a Stephen King-esque horror novel that will be just as good as anything Stephen King puts out.
AI can't do literary effects. Neither can actual writers.
I have the correct plot for a sequel to LIve, Die, Repeat (2014), which badly needs a correct sequel. If they did one on their own they'd screw it up.
In the existing film the woman saves the man from himself; in the sequel he has to save her from herself.
Conveniently it would be from starting over, the way the old film left off.
"We are going to see a lot of articles like this, because it is a fundamental truth that AI is derivative of existing art and cannot create a new piece of art. It can only alter existing art to generate facsimiles of a style."
This is right, I don't see where the "yet" part follows deductively from what it has done so far. This is like the "yet" in making EVs work without a gas engine backup. It assumes a breakthrough that may never come. If we knew that there was a way to do it, it wouldn't be a "breakthrough."
Like those two stories side by side, one clearly relayed a sense of human experience involving the senses, the other kind of glided along above the senses, and did make the scaffolding of a story that a human could use to do some of the scutwork of story writing, I guess. For some reason I am stuck on "The Big Sleep" after reading it a couple of months ago, and it was full of human insight that was tangential to the plot, maybe, but also made the plot far more satisfying.
Not to mention that there are so many restraints punished into the LLMs for badthink and wrongthink, that it probably will never be able to write anything as transgressive as even Huckleberry Finn was in its time. Because if AI were were developed by the powers that be at that time who completely accepted racism, remember "Birth of a Nation" was viewed in the White House, it could not have written a novel as condemnatory of racism as that book was. It will probably only be able to handle themes that the people in power today approve of, which may very easily be found to be utterly misguided by the next zig or zag of history.
@MJB Wolf, people keep forgetting “yet.” Fifty years ago I took graduate courses in AI as a doctoral student. The state of the art back then compared to today is like stone spear points compared to nuclear-tipped ICBMs. Getting from here to “enduring art with heart” will not take anything like another fifty years.
AI can't make art, because AI isn't needed for money laundering. Art has long since lost any assumption of either skill or deeper meaning, it's mostly about the little cards next to the art describing what it means and also any connections the artist has. And I say this as someone who appreciates the early 20th century modern art movement quite a bit.
Literature is a lot different, there's still a kind of expectation of achievement. Though I wouldn't be surprised if a lot of shows and movies these days had a big AI component. Probably too long ago to have made use of AI, but it sure would explain the Star Wars sequels.
My guess is a lot of pop fiction and low expectation writing will benefit a lot from AI. Being more like fan fiction quality, which a lot of people really do just want to read.
I am no medical expert, but I seem to remember reading that the impact of a frontal lobotomy on a patient's ability to live day to day didn't show up immediately. What a lobotomy seemed to do was to make it impossible for the patient to handle novel situations, so as life changed, as it inevitably does, the patient became more and more helpless.
Relying on AI to create our art, as it did in the novel 1984, would be to lobotomize our society. Yeah there are a lot of hacks freeloading off of the reputation of "Art" as a vital way of processing human experience, there are some artists, like Orwell did, digesting events and advancing our understanding of the human experience.
I have made this point before, but the Ancient Greeks allowed all subjects in their plays, and the Romans restricted the allowable themes to those promoting the glorification of Rome, which plays are still being performed thousands of years later and are still being incorporated in our modern literature, which includes film? AI won't be able to do what the Greeks did, ever. There is no "yet."
There has always been a market for bad writing, which I define broadly, to include stuff that is maybe read on flights, but nobody will read in five years except by accident, and I am guessing that AI can write bad fiction already, with a little bit of guidance.
I think MJB Wolf is more likely right. Great art isn't just about technical excellence (or even about technical excellence at all). Great art evokes emotions. Computing is not on course for that no matter how much time you give it.
I admit I don’t know much about AI, but here’s a scenario for consideration for those who do know more about its development:
AI seems to be based on learning from one side. As an example: art - take millions of examples of human-made art and ‘learn’ how to mimic it. Take some human specified inputs and produce an output based on those inputs plus all of the ‘learning’.
Is anyone considering AI ‘learning’ from the critic’s side and using that as feedback? How about training the AI on millions of human critical reviews of the art to ‘learn’ why a piece of art is loved and admired by humans. Then have the AI judge itself: it makes the art based on the inputs, criticizes it based on its learned capabilities of criticism, and iterates until its critical score reduces to some threshold value.
The best reason why AI won't replace humans in literature and in the arts generally was expressed by Michael Costa on the Daily Show: " ... because all great artists have something that AI will never have: horrible parents."
AI will never write as badly as a human who writes badly.
"fundamentally alien"
We'll see about that.
Never mind the AI, ideas are easy. Execution is where the work is. If all this person did was provide an idea, they don't deserve more than a finders fee.
Alien today; who knows about tomorrow.
Yes. Most of the public thinks the value is in the idea. But whatever the idea, a thousand people have had the same idea already. Value is in execution.
Most novels aren't art. Much art isn't art either. What AI could help non-writers and non-artists do is write the novelization that becomes the basis for the eventual film or television series, which is what even the cultural gatekeepers are paying attention to now. Look at the "culture" section of most periodicals with significant circulation, even magazines that had a great influence on literary life in the past. The novel isn't a pressing concern of theirs.
One day, there'll be a market for early AI anything.
Basic fiction was well diagramed by Vonnegut and others, boy meets girl, boy loses girl, boy goes on a journey to find girl, etc. novels are padded with long descriptions of technology, locations, psychology, long scenes of the intricacies of tech, war, fighting, the legal system, trials, and business or markets and such.
AI based on machine learning is about finding relationships and patterns. It could easily deduce the patterns used in thousands of books and come up with a draft, the author could modify the characters and the sex as they wish.
Novels are only the feedstock for the culture machine that is movies. But if you aren't reading novels, you are one step removed from the source of ideas. For instance I am watching a series on Apple TV right now called "Bad Monkey" which I only watched because of Vince Vaughn, whose judgement on material I trust, and stayed to watch because Meredith Hagner plays such a great villain, anyway, it is based on a novel and watches like a novel. But in today's world, the movie is the finished product, but the novel is one piece of art on a grand scale that an artist can execute singlehandedly, and if that artist is hoping for a wider audience, he must wait for the movie people to find it.
That's why most novels today are written as if they were movies and a limited, high production value series like Bad Monkey is the perfect way to bring the complexities of a complete novel to the screen.
@tim maguire, I wish I was young enough to put up a bet with you, but I’m 78 now and I’m thinking 8 - 10 years.
This is the future, and it's probably already here. Instead of the 12 basic plots, or whatever, it could break down the categories even more finely, the the 12,000 basic plots, and the writer could spin a wheel, and be given an outline of a novel, or describe his idea in vague terms, and have a bespoke plot handed to him based on that pattern analysis, then he puts his own life experiences into it to give a sense of authenticity to the writing.
Here is another plot summary even more general, and that applies to most successful commercial fiction. Hero just trying to live his or her life, some event comes about to disrupt his or her life, he is buffeted by those events, taking blow after blow, then the hero finally gains some kind of mastery of the forces at play , and begins to influence events rather than just being victimized by them, and gains increasing control of the situation, eventually winning the battle, getting the girl, getting the guy, saves the planet, whatever.
How about this: a Russian family gets embroiled in the Napoleonic wars. It just needs adding a few details.
I know, I know. The daughter of a Georgia plantation owner has man trouble during and after the civil war. It could be boffo, except for the slaves.
I just finished my re-read of Barth's "The Sot-Weed Factor." No machine could write or appreciate (or deprecate) that.
A Turing test for AI generated "art" versus human created art has yet to be developed. Sure, we can tell AI had a hand in a lot of crap we see online, with a storyline or images that are quite a bit off balance, but that isn't where the issue will be decided. When an unknown artist shown in a gallery is suddenly acclaimed, with paintings selling like hotcakes for millions, who will know if AI had a hand in the production? And if the current art world with its use of art pieces as money laundering vehicles is any indicator, nobody will care.
I'm not feeling it. Frankly, tcrosse, I don't give a damn.
Suggestion: A veterinarian from Glasgow becomes an apprentice to an older vet in Northern England. A race horse is involved.
Four sisters in the North during the Civil War. One marries a tutor, one marries a German professor, one dies, and one marries a neighbor.
A governess marries her employer. He has a mad wife in the attic.
British officer who was declared dead in the American Revolutionary war returns to the tin mines in Cornwall.
I instructed Microsoft Copilot to "Write a 750 word short story in the suspense genre that has elements that stimulate pleasurable fascination and excitement mixed with apprehension."
It returned a short story with lots of words and no plot. In part ...
"At the end of the corridor, she found another door, this one adorned with strange symbols that glowed faintly in the darkness. With a mixture of fear and determination, she pushed it open and stepped inside. The room beyond was a grand ballroom, its chandeliers casting an eerie light on the empty dance floor.
In the center of the room stood a figure, shrouded in shadows. Emma’s breath caught in her throat as the figure turned to face her. It was a woman, her features obscured by a veil, but her eyes glowed with an unnatural light. The whispering voices grew louder, filling Emma’s mind with a cacophony of sound."
A writer takes her 12 year old daughter with her on a romantic trip to a Greek isle with her boyfriend.
"filling Emma's mind with a cacophony of sound."
Better than an earful of caca.
I cannot imagine AI ever creating a novel like "To Kill A Mockingbird." Just don't see it.
That was a "read it in one sitting" book for me, when I was maybe 20, I laid down in bed with it to read myself to sleep, and put in down at something like 10 am the next day. What a great book. Giles Goat Boy wasn't bad either.
OMG, that sounds like a novel I tried to read for a book club I wanted to join, during the pandemic, and it was on-line, and I just couldn't, and I said to myself, "If this is the kind of crap this book club reads, well, I guess I will just stick to posting about my reading on Althouse, from time to time. It was by the author of "The Night Circus" called "The Starless Sea" which was found after you navigated enough of those rooms.
It was much the same for me the first time, if not one-sitting. Catch-22, Gravity's Rainbow, and Earthly Powers were likewise fairly unputdownable.
I confess that I had forgotten almost all the details of Sot-Weed Factor, and recalled mostly the vibe. Giles is another possibility for reinspection.
I re-read C-22 this summer, and have been eyeing Earthly Powers. Not sure about the Pynchon.
Another common one. Young future hero in humble circumstance is taken into a group as a lowling nothing. Educated somewhat cruelly and exposed to the trade. He shows aptitude and masters the business and becomes a magnate or General. The story is flashbacks from a current existential crisis where his old friends help him. The flashbacks explore how he met his friends and how he was trained in the past to meet his current crisis. Also how he rebelled against his trainers and staked out a place as leader of them all. Plus all the girls he has sex with.
I even came up with a catchy title out of the blue: It was the best of times, worst of times: A memoir of the lived experience of one Ishmael the Wanderer
And I don't think it is intended to at this point. AI is trained to find the most likely solution to a given problem. If you consider book quality on a bell curve with To Kill a Mockingbird on the far right extreme, the goal of AI these days would be to produce a book that fails in the middle of the curve, right at the mean.
What needs to happen to get beyond that is the development of some way to quantify the characteristics of generating output that is placed one or two standard deviations above the mean and use that for the basis of training. Crack that nut and patent it and you would be able to buy X from Elon with your pocket change.
I think you are right about the restraints on bad think and wrong think, Imtay. Back in the 80's in law school I took a course on "Law and Literature," and did a presentation defending Twain from the charge of being a racist for writing Huck Finn. I was a little startled by the negative response. No historical context whatsoever. They wanted Jim to be Will Smith or something (the old Will Smith--the one everybody liked).
Yeah, that's the same story, just told out of order using flashbacks due to the presumed low attention span of the expected audience.
I am sure there are folks here that know more than me, but my impression is that AI can swiftly zoom around the pool of information available on the internet and organize it into acceptable forms for human consumption. It can also analyze items found in that pool of information, which might offer new insights to humans. But pushing outside the boundaries of where people have already been? Don't see it. Let's see AI create a novel treatment for prostate cancer. Let's see it teach us how to communicate with dolphins. Let's see it create peace in the Middle East without total eradication of one or more of the contending parties. Let's see it create a new art form.
Well how about an ai novel where Sven from Minnesota joined the first Minnesota infantry during the civil war and was sent to Gettysburg and charged down the hill to the Alabamans on July 2 and was hit on the head and captured and got amnesia and was captured and taken south where he became a racist and built a plantation that made catfish treated with lye called cattafisk that was sold north. His rich children rebelled and took their trust funds to Minneapolis and paid Upton Sinclair to burn the city.
As others have pointed out, sometimes humorously, you need more than a "great idea for a story". Correspondent in 1920s Paris, with an old war wound, socializes with the smart set. USA ambulance driver experiences the battle of Caparetto, is wounded, and falls in love with his nurse. Communist Dynamite expert tries to blow up a fascist Bridge with a guerilla band during SCW.
Young First LT joins Old Destroyer MInesweeper during WW II and deals with a crazy captain. And Private eye in 30s SF, deals with a band of crooks after an expensive bird sculpture.
Its not the ingrediants (sic) its the cook.
Post a Comment